Digital Signal Processing Reference
In-Depth Information
Fig. 1.
The denoising experiment on the standard image:(a)the original image;(b)the noisy
image(
σ
=20);(c) image denoised by BLS-GSM;(d)image denoised by NLM;(e) image
denoised by BM3D; and (f) image denoised by SNIG-NLM
n
Based on a detailed analysis of the denoised results, BLS-GSM generates some
discontinuous information during the course of denoising, and destroys the flatness of
homogeneous area shown in Figure 2 (c). Especially, the effect is revealed around the
edge of hat obviously. Moreover, NLM method usually leads to two drawbacks such
as excessive smoothness of edges, decreasing the contrast of images. Both of them
give rise to the lost of important details. For example, the breach appears on the
mouth of 'Lena', which makes the contour of mouth broken. So the denoised result of
NLM is the worst within these methods on the visual effect, which is also presented
on the table of PSNR displayed subsequently.
Table 1.
PSNR values (dB) obtained by different methods applied on different test images and
different noise level
BLG-
GSM(dB)
BM3
D(dB)
NLM
(dB)
SNIG-
NLM (dB)
Boats
33.58
29.37
26.36
23.74
33.91
29.62
26.78
23.97
32.88
28.61
25.21
22.16
33.70
29.55
26.70
23.85
σ
=10
n
σ
=25
n
σ
=50
n
σ
=100
n
Lena
35.59
31.73
28.64
25.68
35.92
32.07
29.06
25.95
34.33
30.47
27.30
23.75
35.88
31.97
28.95
25.80
σ
=10
n
σ
=25
n
σ
=50
n
σ
=100
n
